Hello World — Why I'm Starting a Blog
On building in public, writing as thinking, and documenting the research journey.
I've been meaning to do this for a while. Research papers are great for communicating finished results, but they strip away the messy parts — the dead ends, the intuitions that didn't pan out, the tricks that actually made things work.
This blog is for the in-between stuff. The things I wish someone had written down when I was getting started.
What I'll write about
The through-line is the self-evolution gap in today's foundation models — the distance between what they can do once trained and what they could do if they could keep learning, critiquing, and improving themselves in deployment. I'll write about:
- Where MLLMs and agentic systems break down on long-horizon, self-corrective reasoning
- Continual and online learning — keeping models current without catastrophic forgetting
- Post-training and alignment recipes (SFT, RL, preference signals) and what actually moves the needle
- Notes on building evaluation benchmarks that measure what we think they measure
- Occasional thoughts on the research process itself
Why now?
Working at Amazon AGI feels like the right inflection point. New problems, new scale, new collaborators, and a daily reminder of how far frontier models still are from systems that can grow with their users. Writing helps me think clearly about what I'm learning.
The act of writing is the act of discovering what you believe. — David Hare
Let's see where this goes.